394 
Since 42250 of such tasks are to be performed every second, 160 processors 
are needed to achieve the task. These 160 processors are divided into 20 
clusters of 8 processors. 
Similarly, deskewing one range is independent from deskewing another 
range, so that the task can be shared among several processors. About 
4 us are necessary for a TMS-320 to perform the cubic interpolation needed 
to compute each output point. Hence, since output rate is 1.5 Megapixels - 
per second, 6 processors are needed for deskewing. 
MCTM must contain at least 256 rows of 4800 points, i.e. 1.2 mega complex 
words. Since a complex word needs 2 bytes, 2.4 Megabytes are necessary. 
In fact, the easiest solution consists in providing three such banks, i.e. 
7.2 Megabytes to allow "overlap" of azimuth processing tasks (5). In the 
same way, RCTM must contain 256 rows of 4800 points. Here, 8 bits real 
numbers are sufficient so that its size is 1.2 Megabyte. But here a flip 
flop mechanism is necessary: While azimuth processors perform look 
summations on one bank, the deskew processors read data from the other 
bank. Hence the RCTM size is 2.4 Megabytes. 
In the proposed implementation (see Fig. 2), each of the 20 clusters of 8 
azimuth processors is linked to the double ring bus through a PBI module. 
Each PBI module has a 480 K bytes Rams, 360 for MTCM ad 120 K for RCTM. 
Hence both MCTM and RCTM are divided among the 20 algorithm processing 
abonnees. The input data flow from the ADC is entered into HSPP via ABI's 
and routed towards MCTM through the double ring bus. 8 ABI's are 
necessary to handle the 38.4 MB/s input data flow. The rate on the bus is 
16.2 MB/s (time stretched). 
The 6 deskew processors are linked to double ring datas through one ABI 
module which reads datas from RCTM. Output data is then directly sent to 
the output coupler through the internal bus of the deskew processor. 
They do not use ring busses. Input ABI's need small Rams (FIFO) for time 
stretching of the input data bursts and to allow some slight traffic 
fluctuations due to the bus mechanism. Deskew processors ABI needs a 
64 Kbyte memory: two 4800 byte buffers used in a flip flop mode for each 
of the 6 deskew processors. 
The azimuth processors needs for every set of 240 FFT's a new set of 
auxilliary data (reference function vector and information for output 
sample selection), which is generated by an external control unit. 
Multiplexing of auxilliary information with input data is performed 
through a dedicated ABI and input coupler. 
The auxilliary data for the deskew processor is generated by the azimuth 
processors and sent along with the data block. 
For the computation of mean Doppler information a selection of input data 
has to be made as follows: 
- selection of one azimuth line out of 8 : 
- selection of 64 consecutive range lines out of 256 (i.e. there are 
gaps of 192 range lines). 
This is be done by connecting the mean Doppler processor via a PBI to the 
bus, which is loaded under control of the input ABI's. The PBI contains a 
corner turn memory of 80 kByte. The output of the mean Doppler processor 
is directly (i.e. not via ring bus) coupled to the external control unit.